[Insight-users] Re: On Metrics of registration

[Luis Ibanez]

Hi Cspl,


The Image Metrics used by the registration framework
share a basic common API. However, depending on the
specific concept they evaluate, metrics will require
additional parameters that users must provide.

MutulaInformation works by randomply sampling pixels
from the images.For this reason this particular metric
has a "SetNumberOfSamples()" method.  The other metrics:
MeanSquares and NormalizedCorrelation do not subsample
the images and henceforth they do not have this method.

The same situation arises with SetFixedImageStandardDeviation
which is used by the MutualInformation class in order
to get a value that will help to estimate the joint
probabilities of gray levels betwen the images.
The othere metrics do not require to estimate joint
probabilities, so they do not have these methods.

If you are modifying the ImageRegistration example
in order to replace the MI metric for another one,
just remove the method calls that are not applicable
to the new metric you want to test.


BTW Just for the record, the pixel sampling performed
by MI is just a speed up of the basic way of computing
the metric. This is the implementation of the Viola &
Wells paper.  The same speed up could be applied to
any of the other metrics that currently visit all the
pixels in the fixed image. A random sampling would
largely reduce the computing time for registration
of metrics like MeanSquares and NormalizedCorrelation.


Please let us know if you find any other problem.


Thanks


   Luis



----------------

====================================================================

Dear Mr.Luis,

  I am working on itkImageRegistration with different metric classes as
metrics for registration.I could get the output with 
itkMutualInformationImageToImageMetric class.But,I am getting error
when i used itkNormalizedCorrelationImageToImageMetric and
itkMeanSquaresImageToImageMetric.I found that some classes like
setFixedImagestandarddeviation,setNumberOfSamples etc are not available
in Normal and meansquares.I am getting problem at startregistration().

I am unable to catch the exception using try and catch.I am getting some
memory exception.I am enclosing  code.